refactor: Break samplers into classes and store them on agent.sampler.*

lib/agent.js

@@ -5,7 +5,7 @@
 
 'use strict'
 
-const AdaptiveSampler = require('./samplers/adaptive-sampler')
+const determineSampler = require('./samplers/determine-sampler')
 const CollectorAPI = require('./collector/api')
 const ServerlessCollector = require('./collector/serverless')
 const DESTINATIONS = require('./config/attribute-filter').DESTINATIONS
@@ -287,12 +287,14 @@ function Agent(config) {
     this.collector,
     this.harvester
   )
-  this.transactionSampler = new AdaptiveSampler({
-    agent: this,
-    serverless: config.serverless_mode.enabled,
-    period: config.sampling_target_period_in_seconds * 1000,
-    target: config.sampling_target
-  })
+
+  // Instantiate sampling decisions.
+  this.sampler = {}
+  // TODO: This could create duplicate AdaptiveSampler instances of the same
+  // `sampling_target`, fix will be addressed in https://github.com/newrelic/node-newrelic/issues/3519.
+  this.sampler.root = determineSampler({ agent: this, config, sampler: 'root' })
+  this.sampler.remoteParentSampled = determineSampler({ agent: this, config, sampler: 'remote_parent_sampled' })
+  this.sampler.remoteParentNotSampled = determineSampler({ agent: this, config, sampler: 'remote_parent_not_sampled' })
 
   this.queries = new QueryTraceAggregator(
     {
@@ -848,10 +850,10 @@ Agent.prototype._listenForConfigChanges = function _listenForConfigChanges() {
     self.txSegmentNormalizer.load.apply(self.txSegmentNormalizer, arguments)
   })
   this.config.on('sampling_target', function updateSamplingTarget(target) {
-    self.transactionSampler.samplingTarget = target
+    self.sampler.root.samplingTarget = target
   })
   this.config.on('sampling_target_period_in_seconds', function updateSamplePeriod(period) {
-    self.transactionSampler.samplingPeriod = period * 1000
+    self.sampler.root.samplingPeriod = period * 1000
   })
   this.config.on('event_harvest_config', function onHarvestConfigReceived(harvestConfig) {
     if (harvestConfig) {

lib/samplers/README.md

@@ -0,0 +1,90 @@
+# Sampling at New Relic
+
+The New Relic agent supports a robust sampling decision making interface. This is a work-in-progress feature.
+
+## Config
+
+Customers configure how they would like their transactions to be sampled under our `distributed_tracing` section in our config. Remember, sampling will only apply if a customer has `distributed_tracing.enabled` set to `true`.
+
+### Types
+
+- A "sampler mode" refers to the following config sections: `distributed_tracing.sampler`, `distributed_tracing.sampler.full_granularity`, and `distributed_tracing.sampler.partial_granularity`. They are defined by the three sections: `root`, `remote_parent_sampled`, and `remote_parent_not_sampled`.
+- A "sampler section" refers to `root`, `remote_parent_sampled`, or `remote_parent_not_sampled` within a particular sampler mode. The config defined at this section, i.e. `SAMPLER_TYPE: SAMPLER_SUBOPTION?`, describes the sampling decision for that particular scenario within that mode.
+  - `root`: This is the main sampler for traces originating from the current service.
+  - `remote_parent_sampled`: The sampler for when the upstream service has sampled the trace.
+  - `remote_parent_not_sampled`: The sampler for when the upstream service has not sampled the trace.
+
+NOTE: `distributed_tracing.sampler` only exists for backward compatiability and may be deprecated in favor of `distributed_tracing.sampler.full_granularity`. For now, `full_granularity` will take precedence over the old path.
+
+### Full Config in Accordance to Spec
+
+```yaml
+...
+(Serverless DT attributes; they may be defined under distributed_tracing instead)
+account_id: string (unset by default, only set when using Serverless Mode)
+trusted_account_key: string (unset by default, only set when using Serverless Mode)
+primary_application_id: string (unset by default, only set when using Serverless Mode)
+...
+distributed_tracing:
+  enabled: boolean (default true)
+  exclude_newrelic_header: boolean (default false)
+  enable_success_metrics (OPTIONAL): boolean (default true, set to false to disable supportability metrics)
+  sampler: (section for sampling config options for different scenarios)
+    adaptive_sampling_target (see note on Sampling Target below)
+    root: (when the trace originates from the current service)
+      ${SAMPLER_TYPE} (See `Sampler Options` below)
+        ${SAMPLER_SUBOPTION}
+    remote_parent_sampled: (when the upstream service has sampled the trace)
+      ${SAMPLER_TYPE} (See `Sampler Options` below)
+        ${SAMPLER_SUBOPTION}
+    remote_parent_not_sampled: (when the upstream service has not sampled the trace)
+      ${SAMPLER_TYPE}
+        ${SAMPLER_SUBOPTION}
+    full_granularity:
+      enabled
+      root: (when the trace originates from the current service)
+        ${SAMPLER_TYPE} (See `Sampler Options` below)
+          ${SAMPLER_SUBOPTION}
+      remote_parent_sampled: (when the upstream service has sampled the trace)
+        ${SAMPLER_TYPE} (See `Sampler Options` below)
+          ${SAMPLER_SUBOPTION}
+      remote_parent_not_sampled: (when the upstream service has not sampled the trace)
+        ${SAMPLER_TYPE}
+          ${SAMPLER_SUBOPTION}
+    partial_granularity:
+      enabled
+      type   ("reduced", "essential", "compact")
+      root: (when the trace originates from the current service)
+        ${SAMPLER_TYPE} (See `Sampler Options` below)
+          ${SAMPLER_SUBOPTION}
+      remote_parent_sampled: (when the upstream service has sampled the trace)
+        ${SAMPLER_TYPE} (See `Sampler Options` below)
+          ${SAMPLER_SUBOPTION}
+      remote_parent_not_sampled: (when the upstream service has not sampled the trace)
+        ${SAMPLER_TYPE}
+          ${SAMPLER_SUBOPTION}
+...
+```
+
+## Solution
+
+There are three sampler modes, each with three sampler sections, resulting in potentially nine different sampling decisions that the agent would have to support. We create a new `Sampler` instance (`AdaptiveSampler`, `AlwaysOnSampler`, `AlwaysOffSampler`, or `TraceIdRatioBasedSampler`, defined in this folder) for each of these sampler modes' sections.
+
+`agent.sampler` would be defined as:
+
+* `agent.sampler.fullGranularity.root`
+* `agent.sampler.fullGranularity.remoteParentSampled`
+* `agent.sampler.fullGranularity.remoteParentNotSampled`
+* `agent.sampler.partialGranularity.root`
+* `agent.sampler.partialGranularity.remoteParentSampled`
+* `agent.sampler.partialGranularity.remoteParentNotSampled`
+
+These fields currently exist (before core tracing was implemented); `agent.sampler.fullGranularity.*` will take precedence over these fields:
+
+* `agent.sampler.root`
+* `agent.sampler.remoteParentSampled`
+* `agent.sampler.remoteParentNotSampled`
+
+These samplers have a `applySamplingDecision({transaction})` function, which `Transaction` calls (in `lib/transaction/index.js`) to update its `sampled` field and therefore its `priority`.
+
+Unlike the other samplers, the `AdaptiveSampler` must share state with other `AdaptiveSamplers` with the same `sampling_target`, which complicates our seperate sampler instances approach. This will be fixed shortly, and this document will be updated to describe that solution.

lib/samplers/adaptive-sampler.js

@@ -5,7 +5,9 @@
 
 'use strict'
 
-class AdaptiveSampler {
+const Sampler = require('./sampler')
+
+class AdaptiveSampler extends Sampler {
   /**
    *
    * @param {object} opts Sampler options.
@@ -15,6 +17,7 @@ class AdaptiveSampler {
    * @param {boolean} opts.serverless - Indicates if the environment is serverless.
    */
   constructor(opts) {
+    super() // no-op
     this._serverless = opts.serverless
     this._seen = 0
     this._sampled = 0
@@ -33,6 +36,10 @@ class AdaptiveSampler {
     }
   }
 
+  toString() {
+    return 'AdaptiveSampler'
+  }
+
   get sampled() {
     return this._sampled
   }
@@ -99,6 +106,25 @@ class AdaptiveSampler {
     return false
   }
 
+  applySamplingDecision({ transaction, tracestate = null }) {
+    if (!transaction) return
+    if (tracestate) {
+      // Explicitly set sampled and priority from tracestate intrinsics if available
+      transaction.sampled = tracestate?.intrinsics ? tracestate.isSampled : null
+      transaction.priority = tracestate?.intrinsics ? tracestate.priority : null
+      return
+    }
+
+    // If a tracestate is not defined, then do our normal priority calculation.
+    // eslint-disable-next-line sonarjs/pseudo-random
+    const initPriority = Math.random()
+    transaction.sampled = this.shouldSample(initPriority)
+    transaction.priority = transaction.sampled ? initPriority + 1 : initPriority
+    // Truncate the priority after potentially modifying it to avoid floating
+    // point errors.
+    transaction.priority = ((transaction.priority * 1e6) | 0) / 1e6
+  }
+
   /**
    * Starts a new sample period after adjusting the sampling statistics.
    */

lib/samplers/always-off-sampler.js

@@ -0,0 +1,16 @@
+/*
+ * Copyright 2025 New Relic Corporation. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+const Sampler = require('./sampler')
+
+class AlwaysOffSampler extends Sampler {
+  applySamplingDecision({ transaction }) {
+    if (!transaction) return
+    transaction.priority = 0
+    transaction.sampled = false
+  }
+}
+
+module.exports = AlwaysOffSampler

lib/samplers/always-on-sampler.js

@@ -0,0 +1,16 @@
+/*
+ * Copyright 2025 New Relic Corporation. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+const Sampler = require('./sampler')
+
+class AlwaysOnSampler extends Sampler {
+  applySamplingDecision({ transaction }) {
+    if (!transaction) return
+    transaction.priority = 2.0
+    transaction.sampled = true
+  }
+}
+
+module.exports = AlwaysOnSampler

lib/samplers/determine-sampler.js

@@ -0,0 +1,42 @@
+/*
+ * Copyright 2025 New Relic Corporation. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+const AdaptiveSampler = require('./adaptive-sampler')
+const AlwaysOffSampler = require('./always-off-sampler')
+const AlwaysOnSampler = require('./always-on-sampler')
+
+/**
+ * A helper function for the Agent to determine what sampler
+ * to use and will log messages about the chosen sampler.
+ * @param {*} params Function parameters.
+ * @param {Agent} params.agent The New Relic agent instance.
+ * @param {object} params.config The entire agent config.
+ * @param {string} params.sampler The sampler type to use: 'root', 'remote_parent_sampled', or 'remote_parent_not_sampled'.
+ * @returns {Sampler} A Sampler e.g. AdaptiveSampler
+ */
+function determineSampler({ agent, config, sampler = 'root' }) {
+  const samplerDefinition = config?.distributed_tracing?.sampler?.[sampler]
+
+  // Always on?
+  if (samplerDefinition === 'always_on') {
+    return new AlwaysOnSampler()
+  }
+
+  // Always off?
+  if (samplerDefinition === 'always_off') {
+    return new AlwaysOffSampler()
+  }
+
+  // Our default ^.^
+  // We'll ignore https://github.com/newrelic/node-newrelic/issues/3519 for now 0.o
+  return new AdaptiveSampler({
+    agent,
+    serverless: config.serverless_mode.enabled,
+    period: config.sampling_target_period_in_seconds * 1000,
+    target: config.sampling_target // getter/setter for distributed_tracing.sampler.adaptive_sampling_target
+  })
+}
+
+module.exports = determineSampler

lib/samplers/sampler.js

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2025 New Relic Corporation. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * @private
+ * @interface
+ */
+class Sampler {
+  /**
+   * Sets `priority` and `sampled` on the transaction
+   * in respect to this sampler's decision.
+   * @param {object} params to make the sampling decision with
+   * @param {Transaction} params.transaction the transaction to update
+   */
+  applySamplingDecision({ transaction }) {
+    throw new Error('must implement applySamplingDecision for %s', transaction)
+  }
+}
+
+module.exports = Sampler